The patient record has outgrown the clinician. What is the approach? Here’s an idea.
Dr. Sedivy discusses how many movies will fit in 11 exabytes of healthcare data. Millions of movies. Most clinical decisions are made on a fraction of it. About 40% of the data lives in documents not designed to be captured in the EHR. Average number of pages for PCP to read in a day. 144,000 pages of data per day for clinician to consume to take care of patient. And it may be on a scanned image or ekg tracing.
Dr. Sedivy describes a new term: CDI: clinical data intelligence layer (a little confusing compared to traditional use of CDI clinical documentation improvement)
Dr. Sedivy describes the problem. Patient satisfaction, quality, revenue, safety, outcomes are all at risk.
Love this metaphor. The icy windshield.
So this is what a theoretical AI tool would need to do. and what Evidently as a company is trying to do, assembling data and moving it toward wisdom.
The research. Have AI look at the chart and write an H/P and compare completeness and accuracy of diagnoses at time of admission.
The outcome. Across all research settings, 4-5 diagnoses missed by the human that was quality-impacting. Eg: atrial fib. Freidrichs ataxia. Hypercoagulable state.
There was a 31 point increase in net EHR experience score per KLAS research methods. And provider satisfaction was 90%. With 60% average time savings on chart review. And an ROI just from HCC capture.
Interesting cases of diagnoses surfaced: afib only within ekg tracing with no interpretation on chart. Also case of recurrent infection. Surfaced the knowledge of implant which had been forgotten by the patient and missed by the clinicians.
How to evaluate new tools in this clinical data intelligence space?
Where this is going: linking chart summary with the ambient scribe so that it goes not only to the clinician but directly into an improved progress note.
And, Evidently has education for responsible AI user training in addition to their tech stack.
Finally we see a clever demo where the AI tool listens to an audio track from The Pitt and answers a complex question about dose of nitrates in a patient with renal failure, hypotension and biventricular failure. The AI suggests nitrates may not be appropriate action especially with RV failure.
Impressive concept, execution and demo.